The 2014 Data Breach Investigations Report


Published on

Company names mentioned herein are the property of, and may be trademarks of, their respective owners.

Published in: Education, Technology, Business
  • Be the first to comment

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

The 2014 Data Breach Investigations Report

  2. 2. 2014 DBIR Contributors (see Appendix C for a detailed list) DEFENS E SECURITY S ERVICE UN IT E D S TAT E S O F A M E R ICA V C D B Malware Analysis & Threat Intelligence ii VERIZON ENTERPRISE SOLUTIONS
  3. 3. CONTENTS INTRODUCTION........................................................................................................................................................................2 2013 YEAR IN REVIEW..........................................................................................................................................................3 VICTIM DEMOGRAPHICS.....................................................................................................................................................5 A DECADE OF DBIR DATA.....................................................................................................................................................7 RESULTS AND ANALYSIS.................................................................................................................................................. 13 POINT-OF-SALE INTRUSIONS...........................................................................................................................16 WEB APP ATTACKS.................................................................................................................................................20 INSIDER AND PRIVILEGE MISUSE...................................................................................................................23 PHYSICAL THEFT AND LOSS..............................................................................................................................27 MISCELLANEOUS ERRORS.................................................................................................................................29 CRIMEWARE...............................................................................................................................................................32 PAYMENT CARD SKIMMERS...............................................................................................................................35 DENIAL OF SERVICE...............................................................................................................................................38 CYBER-ESPIONAGE................................................................................................................................................43 EVERYTHING ELSE.................................................................................................................................................46 CONCLUSION AND SUMMARY RECOMMENDATIONS......................................................................................... 48 APPENDIX A: METHODOLOGY........................................................................................................................................ 51 APPENDIX B: DATA BREACHES AND IDENTITYTHEFT: A CONVOLUTED ISSUE...................................... 53 APPENDIX C: LIST OF CONTRIBUTORS....................................................................................................................... 55 ENDNOTES................................................................................................................................................................................56 Questions? Comments? Brilliant ideas? We want to hear them. Drop us a line at, find us on LinkedIn, or tweet @VZdbir with the hashtag #dbir. 1VERIZON 2014 DATA BREACH INVESTIGATIONS REPORT
  4. 4. INTRODUCTION Welcome to the 2014 Data Breach Investigations Report (DBIR).1 Whether you’re a veteran reader who’s been with us since our initial publication back in 2008 or a newbie to our annual data party, we’re sincerely glad you’re here. We hope that this year’s submission will improve awareness and practice in the field of information security and support critical decisions and operations from the trenches to the boardroom. For DBIR veterans, a cursory look at the table of contents will reveal some significant changes to the report structure you’ve gotten used to in years past. Rather than our signature approach organized around actors, actions, assets, timelines, etc., we’ve created sections around common incident patterns derived directly from the data itself (more on that later). Within each of those patterns, we cover the actors who cause them, the actions they use, assets they target, timelines in which all this took place, and give specific recommendations to thwart them. The drive for change is three-fold: first, we realized that the vast majority of incidents could be placed into one of nine patterns; second, we can (and did) draw a correlation between these incident patterns and industries; and third, we wanted to challenge ourselves to look at the data with a fresh perspective. The ultimate goal is to provide actionable information presented in a way that enables you to hash out the findings and recommendations most relevant to your organization. We all know that data doesn’t grow on trees, and we must express our gratitude to the 50 organizations that contributed to this report, representing public and private entities from around the globe. We’re proud to work with these organizations and feel that what you’re now reading is proof of the benefits of coordinated incident data sharing. For the full list of 2014 DBIR contributors, check out Appendix C. The dataset that underpins the DBIR is comprised of over 63,000 confirmed security incidents — yep, over Sixty-Three Thousand. That rather intimidating number is a by-product of another shift in philosophy with this year’s report; we are no longer restricting our analysis only to confirmed data breaches. This evolution of the DBIR reflects the experience of many security practitioners and executives who know that an incident needn’t result in data exfiltration for it to have a significant impact on the targeted business. So prepare to digest what we hope will be some very delicious data prepared for you this year. The Methodology section, normally found near the beginning of the report, is now in Appendix B. We’ll begin instead with a review of 2013 from the headlines, then provide a few sample demographics to get you oriented with the dataset. The following section — a summary of our 10 years’ of incident data — might just be our favorite. (but please don’t tell the other sections that). We’ll then provide analysis of the aforementioned incident classification patterns and end with some conclusions and a pattern-based security control mapping exercise. So let’s get started! 50CONTRIBUTING GLOBAL ORGANIZATIONS 1,367CONFIRMED DATA BREACHES 63,437SECURITYINCIDENTS 95COUNTRIES REPRESENTED 2 VERIZON ENTERPRISE SOLUTIONS
  5. 5. 2013 YEAR IN REVIEW The year 2013 may be tagged as the “year of the retailer breach,” but a more comprehensive assessment of the InfoSec risk environment shows it was a year of transition from geopolitical attacks to large-scale attacks on payment card systems. 2013 may be remembered as the “year of the retailer breach,” but a comprehensive assessment suggests it was a year of transition from geopolitical attacks to large-scale attacks on payment card systems. JANUARY January saw a series of reports of targeted attacks by what were probably state-sponsored actors. The Red October cyber-espionage campaign was exposed and responsible for targeting government agencies and research institutions globally, but in Russian-speaking countries in particular. Intelligence then connected it to actors using the Elderwood framework, and also a complex series of attacks beginning with a “watering hole” attack on the Council on Foreign Relations web site ( that began on Boxing Day 2012. Meanwhile, the Izz ad-Din al-Qassam Cyber Fighters (QCF) were almost a month into Phase II of Operation Ababil Distributed Denial of Service (DDoS) attacks on U.S. financial services companies. FEBRUARY The segue into February was provided by The New York Times and the Wall Street Journal, with new reports of targeted cyber-espionage. And Sophos reported a new Citadel-based Trojan crafted to attack Point-of-Sale (POS) systems using a Canadian payment card processor. We would soon learn that www. became a watering hole, using a surprise attack on Java late in the month. Most InfoSec professionals well remember February as the month Mandiant (now FireEye) released its superb APT1 report. February was also the start of reports of data breaches from large enterprises, courtesy of the aforementioned iPhoneDevSDK: Facebook, Twitter, Apple, and Microsoft were all victims. Noteworthy retailer POS data breaches were reported by Bashas’ and Sprouts, two discrete grocery chains in the U.S. Southwest. Bit9 reported a data breach that began in July 2012, attacking its code-signing infrastructure. MARCH Fifty million Evernote users remember that March was the month they were forced to change their passwords. On March 20, the Republic of Korea suffered a large-scale cyber-attack that included disk corruption. We remain skeptical that the Cyberbunker-CloudFlare-Spamhaus DoS attack almost broke the Internet at the end of March. Group-IB reported “Dump Memory Grabber” (a.k.a. BlackPOS), a new POS Trojan that would go on to make headlines when news broke of Target Stores’ breach in December. This section is a compilation of the weekly INTSUM lead paragraphs posted to our blog and is 100% based on open source intelligence (OSINT). We maintain a very strong policy against identifying Investigative Response clients, and mentions of organizations in this section in no way imply that we conducted an investigation involving them or that they are among the victims in our dataset. 3VERIZON 2014 DATA BREACH INVESTIGATIONS REPORT
  6. 6. APRIL In April, another U.S. grocery retailer, Schnucks, reported a POS data breach.The Syrian Electronic Army (SEA) did some damage when it hijacked the Associated Press’Twitter account, sending a tweet reporting an explosion at theWhite House and causing a spasm onWall Street. Operation Ababil continued, but OSINT cannot support attributing DoS attacks on several European banks to the QCF. MAY Cyber-espionage continued in May, with reports from QinetiQ and the U.S. Army Corps of Engineers. The SEA hijacked the Twitter accounts of both The Guardian and The Financial Times. A watering hole attack targeted nuclear weapons researchers in the U.S. for cyber-espionage, probably from China. More cyber- espionage campaigns reported in May included Operation Hangover, targeting Pakistan; Safe, targeting Mongolia; and operations by the Sunshop actors against Tibetan activists. The U.S. Department of Justice shut down Liberty Reserve, the go-to bank for cyber-criminals. JUNE Early in June, Raley’s, yet another U.S. grocer with stores in California and Nevada, reported its payment card systems were breached. NetTraveller, a global cyber-espionage campaign targeting diplomats in countries with interests not aligned with China occurred. A day later, The Guardian published the first intelligence leaked by Edward Snowden… and then InfoSec intelligence became the “All-Snowden-All-the- Time” channel. JULY July’s largest retailer data breach was reported by Harbor Freight, a U.S. tool vendor with 445 stores – nearly 200 million customers and we still don’t know how many records were compromised. The QCF initiated Phase IV of Operation Ababil. The SEA breached Viber, Tango, and the Daily Dot. The U.S. Department of Justice indicted four Russians and one Ukrainian for high-profile data breaches, including Heartland and Global Payments. AUGUST In August, the SEA hijacked the Twitter accounts of CNN, The Washington Post, Time Magazine, SocialFlow, and both The New York Times and New York Post. Attendees of the G-8 Summit in St. Petersburg, Russia, were targeted for cyber-espionage by the Calc Team actors. SEPTEMBER In September, Vodafone notified two million customers their personal and financial information had been breached. Espionage reported in September involved the EvilGrab Trojan and separately, the Hidden Lynx actors who seem to engage in both espionage and cybercrime. New intelligence linked the Bit9 attack from February with Operation Deputy Dog, Hidden Lynx, and watering hole attacks on Japanese financial institutions. At the end of the month Brian Krebs began his reports on intelligence extracted from ssndob[dot]ms. The site was home to data stolen from some of America’s largest data brokers: Lexis-Nexis, Kroll, and Dun & Bradstreet. Cryptolocker made its first appearance in September, extorting money from victims that were willing to pay to decrypt their essential files. OCTOBER On October 3, Adobe announced its systems had been breached; eventually 38 million accounts were identified as affected. Intelligence connected this to the ssndob[dot]ms actors. Nordstrom, the luxury U.S. department store, discovered skimmers on some of its cash registers. Two of 2013’s big wins also occurred in October: Dmitry “Paunch” Fedotov, the actor responsible for the Blackhole exploit kit, was arrested in Russia, and Silk Road, an online fraud bazaar, was taken down. NOVEMBER The proverbial calm before the storm, November was fairly quiet. Banking malware evolved with reports of Neverquest and another version of IceIX. BIPS, a major European bitcoin payment processor, was the victim of one of the largest bitcoin heists recorded up to that point in time. DECEMBER The last significant entry under cyber-espionage for 2013 was the targeting of foreign ministries in European countries by Operation Ke3chang. The Washington Post reported its second breach of the year. And then InfoSec intelligence became the “All-Target-All-the-Time” channel. Although the breach of this major U.S. retailer was a little more than half the size of Heartland and three-fourths the size of TJX, it’s vying to become the event for which 2013 will always be remembered.   Questions? Comments? Brilliant ideas? We want to hear them. Drop us a line at, find us on LinkedIn, or tweet @VZdbir with the hashtag #dbir. 4 VERIZON ENTERPRISE SOLUTIONS
  7. 7. VICTIM DEMOGRAPHICS Readers of the DBIR frequently approach us with two important questions. How generally representative are the findings of this report? Are these findings relevant to my organization? To help get you oriented with this year’s report, let’s see what the data has to show us. The 2013 DBIR featured breaches affecting organizations in 27 countries. This year’s report ups that tally by 350%, to 95 distinct countries (Figure 1). All major world regions are represented, and we have more national Computer Security Incident Response Teams (CSIRTs) than ever before. Our ability to compare global trends has never been higher. But it’s not quite that simple. The charter, focus, methods, and data differ so much between CSIRTs that it’s difficult to attribute differences to true variations in the threat environment.2 However, regional blind spots are getting smaller thanks to our growing list of contributors (see Appendix C), and we’re very happy with that. Figure 1. Countries represented in combined caseload Countries represented in combined caseload (in alphabetical order): Afghanistan, Albania, Algeria, Argentina, Armenia, Australia, Austria, Azerbaijan, Bahrain, Belarus, Belgium, Bosnia and Herzegovina, Botswana, Brazil, Brunei Darussalam, Bulgaria, Cambodia, Canada, Chile, China, Colombia, Congo, Croatia, Cyprus, Czech Republic, Denmark, Egypt, Ethiopia, Finland, France, Georgia, Germany, Greece, Hong Kong, Hungary, India, Indonesia, Iran, Islamic Republic of, Iraq, Ireland, Israel, Italy, Japan, Jordan, Kazakhstan, Kenya, Korea, Republic of, Kuwait, Kyrgyzstan, Latvia, Lebanon, Lithuania, Luxembourg, Macedonia, the former Yugoslav Republic of, Malaysia, Mali, Mauritania, Mexico, Moldova, Republic of, Montenegro, Morocco, Mozambique, Nepal, Netherlands, New Zealand, Oman, Pakistan, Palestinian Territory, Occupied, Peru, Philippines, Poland, Portugal, Qatar, Romania, Russian Federation, Saudi Arabia, Singapore, Slovakia, Slovenia, South Africa, Spain, Switzerland, Taiwan, Province of China, Tanzania, United Republic of, Thailand, Turkey, Turkmenistan, Uganda, Ukraine, United Arab Emirates, United Kingdom, United States, Uzbekistan, Vietnam, Virgin Islands. 5VERIZON 2014 DATA BREACH INVESTIGATIONS REPORT
  8. 8. Industry Total Small Large Unknown Accommodation [72] 212 115 34 63 Administrative [56] 16 8 7 1 Agriculture [11] 4 0 3 1 Construction [23] 4 2 0 2 Education [61] 33 2 10 21 Entertainment [71] 20 8 1 11 Finance [52] 856 43 189 624 Healthcare [62] 26 6 1 19 Information [51] 1,132 16 27 1,089 Management [55] 10 1 3 6 Manufacturing [31,32,33] 251 7 33 211 Mining [21] 11 0 8 3 Professional [54] 360 26 10 324 Public [92] 47,479 26 47,074 379 Real Estate [53] 8 4 0 4 Retail [44,45] 467 36 11 420 Trade [42] 4 3 0 1 Transportation [48,49] 27 3 7 17 Utilities [22] 166 2 3 161 Other [81] 27 13 0 14 Unknown 12,324 5,498 4 6,822 Total 63,437 5,819 47,425 10,193 Next, let’s review the different industries and sizes of victim organizations in this year’s dataset (Figure 2). The Public sector’s astronomical count is primarily a result of U.S. agency reporting requirements, which supply a few of our contributors with a vast amount of minor incidents (more on that later), rather than a sign of higher targeting or weak defenses. Figure 3 filters out the minutiae by narrowing the dataset to only those incidents involving confirmed data compromise. Moving beyond the Public sector outlier, both Figure 2 and Figure 3 show demographics relatively similar to prior years. Industry Total Small Large Unknown Accommodation [72] 137 113 21 3 Administrative [56] 7 3 3 1 Construction [23] 2 1 0 1 Education [61] 15 1 9 5 Entertainment [71] 4 3 1 0 Finance [52] 465 24 36 405 Healthcare [62] 7 4 0 3 Information [51] 31 7 6 18 Management [55] 1 1 0 0 Manufacturing [31,32,33] 59 6 12 41 Mining [21] 10 0 7 3 Professional [54] 75 13 5 57 Public [92] 175 16 26 133 Real Estate [53] 4 2 0 2 Retail [44,45] 148 35 11 102 Trade [42] 3 2 0 1 Transportation [48,49] 10 2 4 4 Utilities [22] 80 2 0 78 Other [81] 8 6 0 2 Unknown 126 2 3 121 Total 1,367 243 144 980 We saw some increases where we added new industry-specific contributors, so pieces of the puzzle are filling in. Certain sectors will always skew higher in the victim count given their attractiveness to financially motivated actors — i.e., those that store payment card or other financial data. But even discounting that, we don’t see any industries flying completely under the radar. And that’s the real takeaway here — everyone is vulnerable to some type of event. Even if you think your organization is at low risk for external attacks, there remains the possibility of insider misuse and errors that harm systems and expose data. So, we can’t claim to have unbiased coverage of every type and size of organization on the planet (fingers crossed for next year, though!). But we dare say that the majority of readers will be able to see themselves or something that looks enough like them in this sample. For more information on the NAICS codes [shown above] visit: Small = organizations with less than 1,000 employees, Large = organization with 1,000+ employees Figure 2. Number of security incidents by victim industry and organization size, 2013 dataset Figure 3. Number of security incidents with confirmed data loss by victim industry and organization size, 2013 dataset 6 VERIZON ENTERPRISE SOLUTIONS
  9. 9. A DECADE OF DBIR DATA Long-time readers of this report will know that we’re not very good at maintaining the status quo. The sources of data grow and diversify every year. The focus of our analysis shifts. The way we visualize data and organize results evolves over time. And with the 2014 DBIR, we’re really gonna shake things up. This section attempts to create an “as-comparable-as-possible” set of findings to previous DBIRs. It “only” includes breaches from 2004-2012, plus the 1,367 incidents for which data compromise was confirmed in 2013. While this does make it hard to meaningfully compare trends across time, it has the positive effect of shining light into new and shadowy areas each year. The truth of the matter is that we’re more interested in exploring and learning than churning out the same ‘ol stuff each time just to measure deltas. That said, measuring deltas has value and we know readers appreciate some level of continuity between reports. Thus, this section attempts to create an “as-comparable-as-possible” set of findings to previous DBIRs. It “only” includes breaches from 2004-2012, plus the 1,361 incidents for which data compromise was confirmed in 2013. It’s worth noting that this represents the high mark in ten years of data breaches, and is the first time we’ve crossed 1,000. (Give a round of applause to all those contributors who keep adding fuel to the bonfire.) We began writing a lot of commentary for this section, but then changed our minds. Instead, we’ll churn out some eye candy for you to chew on as long as you like with only a few general observations from us. We began writing a lot of commentary for this section, but changed our minds. Instead, we’ll churn out some eye candy for you to chew on as long as you like, with only a few general observations from us. A BRIEF PRIMER ONVERIS ANDVCDB The Vocabulary for Event Recording and Incident Sharing (VERIS) is designed to provide a common language for describing security incidents in a structured and repeatable manner. It takes the narrative of “who did what to what (or whom) with what result,” and translates it into the kind of data you see in this report. Because we hope to facilitate the tracking and sharing of security incidents, we released VERIS for free public use. Get additional information on the VERIS community site ; the full schema is available on GitHub. Both are good companion references to this report for understanding terminology and context. | Launched in 2013, the VERIS Community Database (VCDB) project enlists the cooperation of volunteers in the security community in an attempt to record all publicly disclosed security incidents in a free and open dataset. We leverage VCDB for a few sections in this report, which are clearly marked. Learn more about VCDB by visiting the website below. 7VERIZON 2014 DATA BREACH INVESTIGATIONS REPORT
  10. 10. Figure 4 depicts the raw count of breaches attributed to external, internal, and partner actors over the 10-year history of our breach data. Figure 5 shows this as a proportion of all breaches and rearranges the categories to highlight exclusivity and overlap among them. It uses a third-degree polynomial trend line to make it nice and smooth, so we can see the basic behavior over time. Together they help answer our primary questions of interest — which actors perpetrate the most breaches and what’s the relative change over time? Since we’re letting the visualizations do most of the talking here, we’ll only make a few observations and leave the rest for homework. • Ten years offers some nice min/max/most likely estimates for you modelers out there. Barring 2006-2008, the overall ratio is relatively stable, especially when you consider the dramatic changes in total breaches and sources in scope each year. • 2007 is the only year showing an insider majority in Figure 4. This is primarily the result of an unusually small Verizon caseload for confirmed breaches and an influx of U.S. Secret Service data from 2006-2008. We essentially crashed two equally sized – but very different – samples together. • That giant dip for external actors in 2012 seen in Figure 4 coincides with an overall drop in breach count that year, mainly due to fewer large, multi-victim POS intrusion sprees targeting SMBs in the dataset. • Thanks to several new partners who focus on insider crimes, the proportional trend line for internal swings up over the last couple years while external turns downward. If we removed the polynomial curving, however, you’d see a positive regression for outsiders and a slightly negative one for insiders. Figure 4. Number of breaches per threat actor category over time 250 500 750 1000 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Partner Internal External Figure 5. Percent of breaches per threat actor category over time 25% 50% 75% 100% 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Partner Internal Collusion External 8 VERIZON ENTERPRISE SOLUTIONS BREACHESVS INCIDENTS? This report uses the following definitions: Incident: A security event that compromises the integrity, confidentiality, or availability of an information asset. Breach: An incident that results in the disclosure or potential exposure of data. Data disclosure: A breach for which it was confirmed that data was actually disclosed (not just exposed) to an unauthorized party.
  11. 11. Two different views indicating how the motives of threat actors have changed over the last five years appear in Figure 6 and 7. The line chart (Figure 6) gives the relative percentage of the top three motives in our dataset, while Figure 7 uses an area plot of total incident counts. • We knew espionage had been rising over the last few years, but the trend line chart surprised us by the degree of convergence with financial motives. Will that continue? • Is this finding merely the result of adding contributors to the DBIR who specialize in espionage, or is money truly diminishing as the prime driver of crime on the Internet? We have an easier time believing the former than the latter, but it certainly makes us want to continue widening our collection of breach data in the future. • The area plot reminds us that money-motivated breaches still outnumber others by a good margin. To borrow from Pink Floyd, most actors still want to “grab that cash with both hands and make a stash.” Figure 8 has the challenging job of exhibiting 10 years of threat actions leading to data breaches. We experimented with alternate ways to visualize this, but thought the simplicity of this chart worked best. Keep in mind that actions aren’t mutually exclusive; several can contribute to an incident (the same goes for actors and assets). • This chart does a superb job underscoring the value of data sharing. You can see the number of breaches and diversity of threats grow as the DBIR transitions from single sample to a meta study. • But it’s not all because of changes in the sample set. Notice how the hacking and malware categories explode upward in 2009 and social tactics begin to climb in 2010. These have parallel stories in the real world (e.g., better automated attack tools and DIY malware kits), and it’s fascinating to see them reflected in the data. 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Malware Hacking Misuse Error Physical Social 200 400 600 800 Figure 8. Number of breaches per threat action category over time Figure 6. Percent of breaches per threat actor motive over time 2013201120102009 2012 25% 50% 75% 100% Ideology/Fun Espionage Financial 2013201120102009 2012 250 500 750 1000 Financial Espionage Ideology/Fun Figure 7. Number of breaches per threat actor motive over time 9VERIZON 2014 DATA BREACH INVESTIGATIONS REPORT
  12. 12. Figure9divesdeeperintothespecificvarietiesofthreatactionsobservedoverthelastfiveyears.Theoveralltoptwentyacrossthefive-yearspanislistedinsuccessivecolumns,and thelinesconnectingcolumnshighlighthoweachactionchangesovertime..Tobehonest,concisecommentaryonthisvisualizationmaybeimpossible.Yes,it’sincrediblybusy,butit’salso incrediblyinformation-dense.Letyoureyesadjustandthenexplorewhateverstrikesyourfancy.Asanexample,followRAMscrapersthroughtheyears.Theystartat#5in2009,dropway downoverthenextfewyearsandthenshootupthechartstothe#4spotin2013.WetalkaboutthatresurgenceinthePOSintrusionssectionofthisreport.LiterallyeveryiteminFigure9has astoryifyoucaretolookforit.Enjoy. Figure9. Top20varietiesofthreatactionsovertime 20102009201220112013 Useofstolencreds[hac]422 Useofstolencreds[hac]203Useofstolencreds[hac]327 Useofstolencreds[hac]84 Useofstolencreds[hac]28 Exportdata[mal]327 Exportdata[mal]183 Exportdata[mal]309 Exportdata[mal]233Exportdata[mal]103Phishing[soc]245 Phishing[soc]181Phishing[soc]62 Phishing[soc]11 Phishing[soc]10 Ramscraper[mal]223 Ramscraper[mal]27 Ramscraper[mal]21 Ramscraper[mal]17 Backdoor[mal]165 Backdoor[mal]209 Backdoor[mal]214 Backdoor[mal]104 Backdoor[mal]267 UseofbackdoororC2[hac]152 UseofbackdoororC2[hac]192 UseofbackdoororC2[hac]237UseofbackdoororC2[hac]202 UseofbackdoororC2[hac]94 Spyware/Keylogger[mal]149 Spyware/Keylogger[mal]215 Spyware/Keylogger[mal]480Spyware/Keylogger[mal]255 Spyware/Keylogger[mal]28Downloader[mal]144 Downloader[mal]181 Downloader[mal]59 Downloader[mal]15 Downloader[mal]13 Capturestoreddata[mal]133 Capturestoreddata[mal]196 Capturestoreddata[mal]58 Capturestoreddata[mal]8 Capturestoreddata[mal]11 C2[mal]119 C2[mal]183 C2[mal]61 C2[mal]15C2[mal]4 SQLi[hac]109 SQLi[hac]25 SQLi[hac]53 SQLi[hac]53SQLi[hac]13 Bruteforce[hac]108 Bruteforce[hac]188 Bruteforce[hac]581 Bruteforce[hac]221 Bruteforce[hac]107 Rootkit[mal]106Rootkit[mal]106 Rootkit[mal]31 Rootkit[mal]0Rootkit[mal]0 Tampering[phy]102 Tampering[phy]56 Tampering[phy]146 Tampering[phy]300 Tampering[phy]22 Disablecontrols[mal]2 Disablecontrols[mal]188Disablecontrols[mal]169 Disablecontrols[mal]102 Disablecontrols[mal]7 Passworddumper[mal]75 Passworddumper[mal]70 Passworddumper[mal]51 Passworddumper[mal]0Passworddumper[mal]0 Privillegeabuse[mis]65 Privillegeabuse[mis]59 Privillegeabuse[mis]33 Privillegeabuse[mis]59Privillegeabuse[mis]18 Scannetwork[mal]62 Scannetwork[mal]101 Scannetwork[mal]2 Scannetwork[mal]38 Scannetwork[mal]1 Adminware[mal]39 Adminware[mal]33 Adminware[mal]28 Adminware[mal]47 Adminware[mal]90 Footprinting[hac]8Footprinting[hac]4Footprinting[hac]2 Footprinting[hac]185 Footprinting[hac]8 Ramscraper[mal]90 10VERIZONENTERPRISESOLUTIONS
  13. 13. Figures 10 and 11 show how the mix of compromised assets has changed over time. It’s useful because it reveals the “footprint” of attackers as they travel through the victim’s environment in search of data. As defenders, it gives us a sense of what may need extra attention or protection. • Servers have typically been on top, probably because attackers know that’s where the data is stored. • User devices have been growing over time, probably because they offer an easy foot in the door. • Media is the only asset category trending down, probably because of an unusually high concentration of (partially-related) cases in 2009 that involved numerous thefts of documents and digital media. • Many ask why the Network category is so low, given that most of these breaches take place over the network. In view here are specific network devices like routers, switches, etc. Malicious traffic definitely passes through those, but they’re not typically compromised during a breach. It would be hard to give proper treatment to a decade of data theft without covering the varieties of data stolen over that time period. Thankfully, Figure 12’s got us covered in that department. • If you compare these trends with those of actor motives from Figure 6 and 7, you’ll see some parallels. Financially-motivated criminals will naturally seek out data that is easily converted to cash, such as bank information and payment cards, while espionage groups target internal corporate data and trade secrets. • The trend for payment card theft is quite fascinating; it rises quickly to a peak in 2010, and then exhibits a negative slope. There’s an uptick in 2013, but it was still the first year in the history of this report where the majority of data breaches did not involve payment cards. • Authentication credentials are useful in both the criminal underground and the shadowy world of the clandestine, and that demand is reflected here. Figure 10. Percent of breaches per asset category over time 2013201120102009 2012 10% 20% 30% 40% 50% Kiosk Person Media User Devices Network Server Bank Payment Credentials Personal Internal Secrets 2004 2006 2005 2007 2009 2008 2010 2011 2013 2012 100 200 300 400 500 2004 2006 2005 2007 2009 2008 2010 2011 2013 2012 100 200 300 400 500 600 600 2004 2006 2005 2007 2009 2008 2010 2011 2013 2012 100 200 300 400 500 2004 2006 2005 2007 2009 2008 2010 2011 2013 2012 100 200 300 400 500 600 600 2004 2006 2005 2007 2009 2008 2010 2011 2013 2012 100 200 300 400 500 2004 2006 2005 2007 2009 2008 2010 2011 2013 2012 100 200 300 400 500 600 600 Figure 12. Breach count by data variety over time Figure 11. Number of breaches per asset category over time 2013201120102009 2012 200 400 600 Kiosk Person Media User Devices Network Server 11VERIZON 2014 DATA BREACH INVESTIGATIONS REPORT
  14. 14. Take a deep, calming breath before diving into this last one; it may result in mental or even bodily harm. In Figure 13, we’re contrasting how long it takes the attacker to compromise an asset with how long it takes the defender to discover this. We chose to peg this on “days” to keep things simple and stark (one might also add “sad” to that alliteration). • Ignore the behavior of the lines for a minute and focus on the wide gap between percentages for the two phases. It smacks us with the fact that the bad guys seldom need days to get their job done, while the good guys rarely manage to get theirs done in a month of Sundays. • The trend lines follow that initial smack with a roundhouse kick to the head. They plainly show that attackers are getting better/faster at what they do at a higher rate than defenders are improving their trade. This doesn’t scale well, people. • We thought about superimposing “total spending on network monitoring,” “number of security products on the market,” and “number of Certified Information Systems Security Professionals (CISSPs) in the workplace,” but we were concerned it would result in much self-inflicted harm within the security community. And we’d much rather you guys and gals stick around and help us fix this. Having dealt that last blow regarding timelines, readers familiar with the traditional flow of the DBIR may expectantly hear that Mortal Kombat imperative of “Finish Him!” in their heads as we head into discussion of breach discovery methods. But there will be no triumphant “Fatality!” announcement here; we’re going to show mercy instead and end on a positive note. • We’re thrilled to see that internal discoveries outnumber external fraud detection for the first time in DBIR history! • It’s great that law enforcement is steadily getting better and better at detecting breaches and notifying victims! • Unrelated third parties, like CSIRTs and threat researchers, are quickly rising as an important and prominent way that victims — especially espionage victims — come to learn about breaches. Keep up the good work, folks; we’re making a dent! We hope you enjoyed this little ten-year trip down memory lane as much as we did. This small band of geeks is grateful to Verizon for allowing us to spend so much time in our playground of breach information. We’re also grateful to the many organizations that have participated in making it possible; without your contributions the data would have gotten stale years ago. And finally, thanks to all you readers out there who download this document and consider these trends as you fight the good fight of protecting information and customers. May the next ten years find us all on the winning side of that battle. Figure 13. Percent of breaches where time to compromise (red)/time to discovery (blue) was days or less 2004 2006 2005 2007 2009 2008 2010 2011 2013 2012 25% 50% 75% 100% Time to compromise Time to discovery Figure 14. Breach discovery methods over time Internal Fraud Detection Law Enforcement Third Party 2004 2006 2005 2007 2009 2008 2010 2011 2013 2012 20% 40% 60% 80% Questions? Comments? Brilliant ideas? We want to hear them. Drop us a line at, find us on LinkedIn, or tweet @ VZdbir with the hashtag #dbir. 12 VERIZON ENTERPRISE SOLUTIONS
  15. 15. RESULTS AND ANALYSIS he seeds of our approach to the 2014 DBIR began to grow during the final phase of drafting the 2013 report. When trying to present statistics around threat actions in a simple and meaningful way, we noticed certain combinations of actors, actions, and assets frequently occurring together within an incident scenario. We gave names to three of these and included some “scratch paper” calculations showing they collectively described 68% of the entire dataset (Figure 15). Production deadlines prevented further exploration into that phenomenon, so we left readers with this thought: “We may be able to reduce the majority of attacks by focusing on a handful of attack patterns.” But as soon as the 2013 DBIR was put to bed, we returned to the notion of incident patterns and began studying our dataset from a very new perspective with a new set of techniques. Now, fast forward to the 2014 DBIR. We have more incidents, more sources, and more variation than ever before – and trying to approach tens of thousands of incidents using the same techniques simply won’t cut it. Not only would the dominant incident characteristics drown out the subtleties of the less frequent varieties, but we cannot continue to study those characteristics as though they occur in isolation. A list of the top 20 actions is helpful, but even more helpful is an accounting of the actors that perform them, other actions used in combination with them, and the assets they tend to target. To reel in that prize, we’re going to need a bigger boat. Full throttle, Mr. Hooper! And that brings us back to recurring combinations of actors, actions, assets, and attributes, or more formally, incident classification patterns. In order to expose these latent patterns in the data, we applied a statistical clustering technique (the bigger boat) by creating a matrix aggregating incidents within each of the common VERIS enumerations and calculating the numeric “distance” between them. This enabled us to find clusters, or patterns, of strongly related VERIS enumerations within the incident dataset. “Strongly related” here essentially means they often occur together in the same incidents and are distinct in some way from other combinations. The first time through, we tossed everything in and looked at the clustering (of the hierarchical type if you’re into that) of VERIS enumerations. Some clusters were obvious, like the social action of phishing with the social vector of email. However, we were looking for clusters that describe comprehensive incident classifications rather than just frequent pairings. For example, incidents involving physical tampering of ATMs by organized criminal groups to steal payment cards stood out like a Wookie among Ewoks. So we labeled that pattern “skimmers,” removed the matching incidents, and reran the cluster analysis on the remaining incidents to look for the next pattern. In the end, we identified nine patterns that together describe 94% of the confirmed data breaches we collected in 2013. Nine out of ten of all breaches can be described by nine basic patterns. But (using our best infomercial voice) that’s not all! When we apply the same method to the last three years of breaches, 95% can be described by those same nine patterns. Figure 15. Scratch paper calculations from the 2013 DBIR for commonly- observed incident patterns 111 POS smash-and-grab 190 Physical ATM + 120 Assured Penetration Technique 421 ÷ 621 Total Breaches 68% 13VERIZON 2014 DATA BREACH INVESTIGATIONS REPORT
  16. 16. But wait — there’s more! Act now, and we’ll throw in all security incidents — not just breaches — from all partners and the VERIS Community Database (VCDB) over the last ten years — for free! Yes, all for the same price of nine patterns, you can describe 92% of 100K+ security incidents! Remember that promise from last year — “We may be able to reduce the majority of attacks by focusing on a handful of attack patterns?” Consider it fulfilled. To us, this approach shows extreme promise as a way to drastically simplify the seemingly endless array of threats we must deal with to protect information assets. We dig into each incident pattern in the following sections, but you can see from Figure 16 that POS intrusions, web app attacks, cyber- espionage, and card skimmers are among the top concerns when we focus on data disclosure. However, it’s not enough to just identify and count the patterns as a whole. Figure 16. Frequency of incident classification patterns Everything else POS Intrusions Cyber-espionage Web App Attacks Insider Misuse Crimeware Miscellaneous Errors Card Skimmers Physical Theft/Loss DoS Attacks 2013 breaches, n=1,367 14% 6% 22% 35% 8% 4% 2% 9% <1% <1% 12% <1% 1% 6% 18% 20% 25% <1% 2013 incidents, n=63,437 14% 3% 31% 15% 21% 8% 4% 1% 14% 5% 2011-2013 breaches, n=2,861 1% <1% Figure 17. Number of selected incident classification patterns over time 250 500 750 1,000 2013201120102009 2012 Card Skimmers Cyber- espionage POS Intrusions Web App Attacks Insider Misuse Figure 18. Percent of selected incident classification patterns over time 2013201120102009 2012 25% 50% 75% 100% Card Skimmers Cyber-espionage POS Intrusions Web App Attacks Insider Misuse 14 VERIZON ENTERPRISE SOLUTIONS
  17. 17. Obviously not every organization needs to focus on point of sale attacks. To make the analysis actionable, we pulled all incidents within each industry and then applied the patterns to create the work of art that is Figure 19. It shows the proportion of incidents within each industry represented by the nine patterns over the last three years. In order to use Figure 19, identify your industry in the left hand column. Refer to the NAICS website if you’re unsure where your organization fits. The percentages are relative to each industry. For example, 10% of all Retail incidents fall within the “web app attack”. The coloring should help you quickly identify “hot spots” for your industry and/or discern differing threat profiles across multiple industries. Before continuing on to the detailed discussion of each pattern (which appear in order according to Figure 18), you may want to study Figure 19. Look up the industry (or industries) that matter to you, identify which patterns are most relevant, and pay special attention to those sections in the report (you’ll still want to read the whole thing, of course). For those curious about how these incident patterns trend over time, we’ve retrofitted them to pre-2013 data to produce Figures 17 and 18. We’ve heard the (constructive) criticism from some of you noting that it’s difficult to pick out exactly which findings from the DBIR apply to your organization, and we spent a lot of time figuring out how to address that. We hope you’ll agree this is a step in the right direction, not only for this report, but also for threat analysis and decision support in general. Figure 19. Frequency of incident classification patterns per victim industry For more information on the NAICS codes [shown above] visit: INDUSTRY POS INTRUS- ION WEB APP ATTACK INSIDER MISUSE THEFT/ LOSS MISC. ERROR CRIME- WARE PAYMENT CARD SKIMMER DENIAL OF SERVICE CYBER ESPION- AGE EVERY- THING ELSE Accommodation [72] 75% 1% 8% 1% 1% 1% <1% 10% 4% Administrative [56] 8% 27% 12% 43% 1% 1% 1% 7% Construction [23] 7% 13% 13% 7% 33% 13% 13% Education [61] <1% 19% 8% 15% 20% 6% <1% 6% 2% 22% Entertainment [71] 7% 22% 10% 7% 12% 2% 2% 32% 5% Finance [52] <1% 27% 7% 3% 5% 4% 22% 26% <1% 6% Healthcare [62] 9% 3% 15% 46% 12% 3% <1% 2% <1% 10% Information [51] <1% 41% 1% 1% 1% 31% <1% 9% 1% 16% Management [55] 11% 6% 6% 6% 11% 44% 11% 6% Manufacturing [31,32,33] 14% 8% 4% 2% 9% 24% 30% 9% Mining [21] 25% 10% 5% 5% 5% 5% 40% 5% Professional [54] <1% 9% 6% 4% 3% 3% 37% 29% 8% Public [92] <1% 24% 19% 34% 21% <1% <1% 2% Real Estate [53] 10% 37% 13% 20% 7% 3% 10% Retail [44,45] 31% 10% 4% 2% 2% 2% 6% 33% <1% 10% Trade [42] 6% 30% 6% 6% 9% 9% 3% 3% 27% Transportation [48,49] 15% 16% 7% 6% 15% 5% 3% 24% 8% Utilities [22] 38% 3% 1% 2% 31% 14% 7% 3% Other [81] 1% 29% 13% 13% 10% 3% 9% 6% 17% 15VERIZON 2014 DATA BREACH INVESTIGATIONS REPORT
  18. 18. POINT-OF-SALE INTRUSIONS WEBAPP ATTACKS INSIDERAND PRIVILEGEMISUSE PHYSICALTHEFT ANDLOSS MISCELLANEOUS ERRORS CRIMEWARE PAYMENTCARD SKIMMERS CYBER- ESPIONAGE DOS ATTACKS EVERYTHING ELSE POINT-OF-SALE (POS) INTRUSIONS The industries most commonly affected by POS intrusions are of no surprise: restaurants, hotels, grocery stores, and other brick-and-mortar retailers are all potential targets. Recent highly publicized breaches of several large retailers have brought POS compromises to the forefront. But at the risk of getting all security-hipster on you — we’ve been talking about this for years. In fact, this is the main cause of the large dip in 2012 seen in many of the “over time” charts in this report. We were writing about RAM scrapers before anyone heard of them and we’re quite frankly not all that into them anymore because they’ve sold out and gone mainstream. Jokes aside, while POS hacks are getting more press recently, they really have been going on for years and we really have talked quite a bit about them in previous DBIRs. The media frenzy makes quite a splash, but from a frequency standpoint, this largely remains a small-and-medium business issue. Focusing too much on outliers and headlines can reflect cognitive bias. For instance, some may be surprised that the number of POS attacks in 2012 and 2013 is substantially lower than the number recorded in 2010 and 2011 (despite having ten times more contributors in the latter years). Figure 20 reminds us that our understanding of risk should always come back to the data, not what makes good headlines and marketing fodder. From an attack pattern standpoint, the most simplistic narrative is as follows: compromise the POS device, install malware to collect magnetic stripe data in process, retrieve data, and cash in. All of these attacks share financial gain as a motive, and most can be conclusively attributed (and the rest most likely as well) to organized criminal groups operating out of Eastern Europe.3 Such groups are very efficient at what they do; they eat POSs like yours for breakfast, then wash ‘em down with a shot of vodka. While the majority of these cases look very much alike, the steps taken to compromise the point-of-sale environment offer some interesting variations. AT A GLANCE Description Remote attacks against the environments where retail transactions are conducted, specifically where card-present purchases are made. Crimes involving tampering with or swapping out devices are covered in the Skimming pattern. Top industries Accommodation and Food Services, Retail Frequency 198 total incidents 198 with confirmed data disclosure Key findings Given recent headlines, some may be surprised to find that POS intrusions are trending down over the last several years. That’s mainly because we’ve seen comparatively fewer attack sprees involving numerous small franchises. Brute forcing remote access connections to POS still leads as the primary intrusion vector. A resurgence of RAM scraping malware is the most prominent tactical development in 2013. We know many of you will come to this section hoping to find all the particulars and dirty laundry of a certain breach involving a major U.S. retailer in late 2013. Prepare to be disappointed; we don’t name victims in this report nor do we divulge client-specific information on any breaches handled by any of the DBIR contributors. If you want up-to- the-minute news on particular breaches, you’d best look elsewhere. As a consolation prize, however, we hope you’ll accept our overall analysis of two hundred POS intrusions that occurred in 2013, along with recommendations on how you can avoid increasing that number in 2014. Figure 20. Comparison of POS Intrusions and Web App Attacks patterns, 2011-2013 2013201120102009 2012 20% 40% 60% Web App Attacks POS Intrusions POINT-OF-SALE INTRUSIONS 16 VERIZON ENTERPRISE SOLUTIONS
  19. 19. POINT-OF-SALE INTRUSIONS WEBAPP ATTACKS INSIDERAND PRIVILEGEMISUSE PHYSICALTHEFT ANDLOSS MISCELLANEOUS ERRORS CRIMEWARE PAYMENTCARD SKIMMERS CYBER- ESPIONAGE DOS ATTACKS EVERYTHING ELSE Let’s start with the most frequent scenario, which affects small businesses that may or may not realize just how lucrative a target they are. This event chain begins with the compromise of the POS device with little to no legwork; the devices are open to the entire Internet and, to make matters worse, protected with weak or default passwords (and sometimes no passwords). The top three threat actions tell the story rather well (Figure 21). The perpetrators scan the Internet for open remote-access ports and if the script identifies a device as a point of sale, it issues likely credentials (Brute force) to access the device. They then install malware (RAM scraper) to collect and exfiltrate (Export data) payment card information. One finding that intrigued us is the renaissance of RAM scraping malware as the primary tool used to capture data. RAM scrapers allow payment card data to be grabbed while processed in memory (where it is unencrypted) rather than when stored on disk or in transit across the network (where it is (ostensibly) encrypted). It’s interesting, but not necessarily surprising, that RAM scraping has usurped keyloggers as the most common malware functionality associated with POS compromises. One could theorize that keyloggers (most of which were common varieties such as Perfect Keylogger and Artemis) are more easily spotted than the memory-scraping code we witnessed in this data set. Or perhaps the RAM scrapers, which hook into specific processes of the POS software, simply do the job better and more efficiently. In years past, we analyzed attack sprees that spanned multiple victims with no association with each other beyond the use of truly awful passwords. This report features almost 200 incidents, but in prior years we saw over 200 victims for one criminal group. The two biggest sprees in our 2013 dataset, one involving several franchisees of the same company, and the other affecting multiple corporations, are a bit different, and lead us to our second common scenario: the use of stolen vendor credentials. In one case the credentials stolen belonged to a point-of-sale vendor and were compromised via Zeus malware infecting the vendor’s systems. The big problem among these was that the same password was used for all organizations managed by the vendor. Once it was stolen, it essentially became a default password and the attackers also gained knowledge of the customer base. Armed with this information, the familiar modus operandi of installing malicious code that captured and transmitted the desired data began. While not as common as the simpler POS intrusions, our dataset does include several incidents from the first quarter of 2013 that feature a compromise at a corporate location, leading to widespread compromise of individual locations and malicious code installations across a multitude of stores. Some cases have begun with a store compromise that led to penetration of the corporate network, but the hub-and-spoke architecture allowed for efficient traversal of the network and the impact of the compromise was magnified regardless of where “device 0” was located. Figure 21. Top 10 threat action varieties within POS Intrusions (n=196) RAM scraper [mal] Export data [mal] Brute force [hac] Useofstolencreds[hac] Offline cracking [hac] Use of backdoor or C2 [hac] Spyware/Keylogger[mal] Backdoor [mal] Misconfiguration[err] Phishing [soc] 37% 8% 2% 2% 1% <1% <1% 85% 79% 50% 38% Bruteforce Use of stolen creds Unknown Offline cracking UseofbackdoororC2 SQLi 2% <1% 9% 9% 53% Figure 22. Hacking variety within POS Intrusions (n=187) 35% 3rd party desktop Desktop sharing Backdoor or C2 Physical access VNP Webapplication <1% <1% 9% 1% Command hell 1% 55% Figure 23. Hacking vector within POS Intrusions (n=187) POINT-OF-SALE INTRUSIONS 17VERIZON 2014 DATA BREACH INVESTIGATIONS REPORT
  20. 20. POINT-OF-SALE INTRUSIONS WEBAPP ATTACKS INSIDERAND PRIVILEGEMISUSE PHYSICALTHEFT ANDLOSS MISCELLANEOUS ERRORS CRIMEWARE PAYMENTCARD SKIMMERS CYBER- ESPIONAGE DOS ATTACKS EVERYTHING ELSE Regardless of how large the victim organization was or which methods were used to steal payment card information, there is another commonality shared in 99% of the cases: someone else told the victim they had suffered a breach. This is no different than in years past, and we continue to see notification by law enforcement and fraud detection as the most common discovery methods. In many cases, investigations into breaches will uncover other victims, which explains why law enforcement is the top method of discovery and the top contributor of POS intrusions in our dataset. Long story short, we’re still discovering payment card breaches only after the criminals begin using their ill-gotten gains for fraud and other illicit purposes. The timelines in Figure 25 reinforce both the compromise vectors and the discovery methods. Entry is often extremely quick, as one would expect when exploiting stolen or weak passwords. Most often it takes weeks to discover, and that’s based entirely on when the criminals want to start cashing in on their bounty. Figure 25. Timespan of events within POS Intrusions Discoveryn=178Compromisen=169Exfilterationn=169 Hours Days Weeks Months Years Never Seconds Minutes 0% 0% 0% 85% 0% 13% 51% 36% 1% 1% 1% 1% 1% 21% 11% 11% 1% 1% 0% 0%0% 0% 88% 5% Figure 24. Top 5 discovery methods for POS Intrusions (n=197) 1%All Internal Ext - law enforcement Ext - customer Ext - fraud detection Int - NIDS Int-reportedbyuser All External 14% <1% <1% 11% 75% 99% BOTNET MITIGATION: AN INCENTIVE PROBLEM According to the NHTCU, the impact of botnets — the Swiss Army knife of cybercriminals ­— remained high in 2013. Furthermore, they note an apparent incentive problem when it comes to mitigating these crafty menaces. Since the impact of a botnet is often spread around the globe, federal authorities aren’t always able to amass resources to fight it on a national level. While the total damage of such a botnet might be large, specific countries only deal with a small part of these damages. The initial costs for fighting such a botnet don’t seem to outweigh the benefits of its takedown. Nevertheless, the NHTCU continue to fight botnets. In February of 2013, public broadcaster NOS presented findings on part of a dropzone of the so-called Pobelka botnet. After an online checking tool was made available, 500,000 people checked to see if their machines had (at some time) been infected; of that group, 23,000 self- identified as victims. By then, the dropzone had been examined for correlations with a 2012 malware outbreak that had prompted a criminal investigation. Sixteen organizations within the vital infrastructure were informed of being infected, and relevant infected IP addresses had been communicated to the respective ISPs. POINT-OF-SALE INTRUSIONS 18 VERIZON ENTERPRISE SOLUTIONS
  21. 21. POINT-OF-SALE INTRUSIONS WEBAPP ATTACKS INSIDERAND PRIVILEGEMISUSE PHYSICALTHEFT ANDLOSS MISCELLANEOUS ERRORS CRIMEWARE PAYMENTCARD SKIMMERS CYBER- ESPIONAGE DOS ATTACKS EVERYTHING ELSE REFLECTIONS AFTERTHE EC3’S FIRSTYEAR IN OPERATION Last year’s DBIR featured an appendix from Troels Oerting, Assistant Director of the European Cybercrime Centre (EC3), discussing the plans and priorities of the newly established division of Europol. Law enforcement agencies play a critical role in this report, and it’s not often we get to see them in their formative stages. Thus, we thought it would be interesting to include some reflections on EC3’s first year of operations. The job of the EC3 isn’t a small one: it serves 28 European Union (EU) member states and dozens of countries, and coordinates protection for 500 million citizens, almost three-quarters of whom have Internet access. In terms of operations, the EC3 prioritizes four areas: cyber intelligence, intrusion, online fraud, and child sexual abuse. As with any new venture, much of the first year focused on building infrastructure and capabilities to fulfill these priorities. Secure network connections to EU and non-EU partners were rolled out, as well as centralized forensic analysis environments and tools. The EC3 trained more than 100 law enforcement experts all over the EU in cyber investigation, tools, and obtaining forensic evidence. It built a new central forensic laboratory to assist member state colleagues in obtaining evidence. It distributed alerts, intelligence notifications, and threat assessments to stakeholders. Memorandums of understanding (MoUs) were signed with key private stakeholders, and a new Advisory Group consisting of experts outside the law enforcement community was established (Verizon is happy to be among them). Trends observed by the EC3 across member states in 2013 include substantial increases in intrusions, malware, phishing, grooming, DDoS, espionage, and botnet activity. It also reports a boom in criminal infrastructure on the darknet, growth in malware affecting mobile devices, and wider distribution of malware from cloud services. In combating these trends, the EC3 has prioritized identifying criminal network operations and cases, with the potential for major and lasting impact. RECOMMENDED CONTROLS FOR ALL COMPANIES The shared vector for the major scenarios is third-party remote- access software (e.g., PCAnywhere, LogMeIn). The security of these products is not the issue here. It just happens that we often find them implemented in a very insecure manner. Restrict remote access Limit any remote access into POS systems by your third-party management vendor, and have serious business discussions regarding how and when they will perform their duties. Enforce password policies Make absolutely sure all passwords used for remote access to POS systems are not factory defaults, the name of the POS vendor, dictionary words, or otherwise weak. If a third party handles this, require (and verify) that this is done, and that they do not use the same password for other customers. “S” is for “Sale,” not “Social.” Do not browse the web, email, use social media, play games, or do anything other than POS-related activities on POS systems. Deploy AV Install and maintain anti-virus software on POS systems. Bottom line: Make it difficult for miscreants to log into a device that accepts the most targeted piece of information for financially motived criminals. FOR LARGE/MULTI-STORE COMPANIES Larger, multi-store companies and franchises should consider a couple of additional recommendations to limit the impact of a single-location breach and prevent a mass compromise. Debunk the flat network theory Review the interconnectivity between stores and central locations and treat them as semi-trusted connections. Segment the POS environment from the corporate network. Look for suspicious network activity Monitor network traffic to and from the POS network. There should be a normalized traffic pattern, and while easier said than done, anomalous traffic must be identified and investigated. Use two-factor authentication Stronger passwords would cut out a huge chunk of the problem, but larger organizations should also consider multiple factors to authenticate third-party and internal users. POINT-OF-SALE INTRUSIONS 19VERIZON 2014 DATA BREACH INVESTIGATIONS REPORT
  22. 22. POINT-OF-SALE INTRUSIONS WEBAPP ATTACKS INSIDERAND PRIVILEGEMISUSE PHYSICALTHEFT ANDLOSS MISCELLANEOUS ERRORS CRIMEWARE PAYMENTCARD SKIMMERS CYBER- ESPIONAGE DOS ATTACKS EVERYTHING ELSE WEB APP ATTACKS There’s no question about it – the variety and combination of techniques available to attackers make defending web applications a complex task. Regrettably, our discussion of this complexity is hampered by the level of detail provided on these incidents. Unless a forensics investigation was performed (a small subset of the overall dataset), the specific techniques utilized went largely unreported or were recorded with broad categorizations. While we have enough material to discuss web application data breaches at a high level, our ability to draw conclusions drops as we dig further into the details (which often aren’t there). Greed takes a back seat to ideology when it comes to web app attacks in the 2013 dataset. Just under two out of every three web app attacks were attributable to activist groups driven by ideology and lulz; just under one out of three came by the hand of financially motivated actors; with the small remainder linked to espionage. After some slicing and dicing we found some very distinct sub-patterns divided along these motives. The financial and ideological attacks deserve unique discussion since the treatment for each may be slightly different. While attacks perpetrated by those motivated by espionage are certainly relevant, discussion of these is taken up in the “Espionage” section. FINANCIALLY MOTIVATED ATTACKS Financially motivated attackers are hyper-focused on gaining access to the money, so it follows that their two primary target industries are the financial and retail industries (where data that easily converts to money is abundant and, all too often, accessible). Within the financial industry, they focus on gaining access to the user interface of the web (banking) application more so than exploiting the web application itself, because the application grants logical access to the money. This means they target user credentials and simply use the web applications protected with a single factor (password) as the conduit to their goal. These could have been included in the section on crimeware (and some did slip through cracks in the algorithm to land there), but the use of web applications as a vector of attack causes them to show up here. The tactics used by attackers are all the usual suspects: a) phishing techniques to either trick the user into supplying credentials or installing malware onto the client system, b) the old stand-by of brute force password guessing, and c) rarer cases of targeting the application through SQL injection or other application-level attacks as a means to retrieve credentials, bypass the authentication, or otherwise target the user-management system. When attribution is possible, the majority of external attackers utilizing stolen credentials somewhere along the attack chain hail from Eastern Europe. Within the retail industry, we see a slightly different focus. The primary aim is payment card information (targeted in 95% of the incidents), which is often accessible simply by exploiting the web application. Social actions (such as phishing) are mostly non- existent, most likely because exploiting vulnerabilities inherent in web applications works plenty well enough. SQL injection was leveraged in 27 of the 34 (80%) attacks against web applications in the retail industry, followed by techniques to install and use web shells (remote file inclusion, etc.) in five of the 34. AT A GLANCE Description Any incident in which a web application was the vector of attack. This includes exploits of code- level vulnerabilities in the application as well as thwarting authentication mechanisms. Top industries Information, Utilities, Manufacturing, Retail Frequency 3,937 total incidents 490 with confirmed data disclosure Key findings Web applications remain the proverbial punching bag of the Internet. They’re beaten in one of two ways: by exploiting a weakness in the application (typically inadequate input validation), or by using stolen credentials to impersonate a valid user. Many of the attacks in our 2013 dataset targeted off-the-shelf content management systems (e.g., Joomla!, Wordpress, or Drupal) to gain control of servers for use in DDoS campaigns. Ideology/Fun Financial Espionage 65% 33% 2% Figure 26. External actor motives within Web App Attacks (n=1,126) WEBAPP ATTACKS 20 VERIZON ENTERPRISE SOLUTIONS
  23. 23. POINT-OF-SALE INTRUSIONS WEBAPP ATTACKS INSIDERAND PRIVILEGEMISUSE PHYSICALTHEFT ANDLOSS MISCELLANEOUS ERRORS CRIMEWARE PAYMENTCARD SKIMMERS CYBER- ESPIONAGE DOS ATTACKS EVERYTHING ELSE IDEOLOGICALLY MOTIVATED ATTACKS: Ideology represents the largest identified portion of motives for web application attacks, and the actors also tend to be the most geographically diverse. 74% focus on tried and true exploits targeting, above all else, unvalidated inputs in executed code. Nowhere is this exploited on a larger scale than Content Management Systems (CMS) such as Joomla!, Drupal, and WordPress, and even then, more in the added plugins than the core CMS code itself. Ideological actors (whether their motivation is social, political, or just for plain fun) are less concerned about getting at the crown jewels than they are about getting a platform (in all senses of the word) to stand on. With that in mind, it’s not surprising that we see two types of results from ideological attackers going after a web server: defacements to send a message or hijacking the server to attack (including by DDoS) other victims. This focus on opportunistically owning just the web server becomes plain when looking at the assets compromised in the attack. The web server was the only asset recorded in nearly all incidents attributable to ideological motives. The actors didn’t appear to be interested in pushing deeper and wider into the network. This result may be the product of simply not reporting those secondary components of the incident — so don’t take this as advice to only focus on the web server — but it is logical and a point of contrast to other types of attacks in our dataset. DISCOVERY METHODS AND TIMELINE When the actor is financially motivated and the discovery method is recorded, we see a leading notification method that we don’t see anywhere else: customers. Perhaps customers notice the fraudulent activity before anyone else, but something is definitely tipping them off before any internal mechanism. With all internal discovery methods combined, only 9% of victims discovered data breaches of their own accord. Discovery method looks a little bleaker for activists. 99% of the notifications were external parties (primarily CSIRTs) contacting victims to let them know their hosts were involved in other attacks. This is heavily influenced by ideological attackers quietly using the platform to attack others rather than, for instance, simple defacements (which are rare in the dataset). Even though the timeline data is a little sparse, it paints the picture of quick entry with 60% of the initial compromises occurring within minutes or less. This reflects the highly repetitive CMS exploits in this pattern; if it works, it works quickly. Just over 85% of the incidents are discovered in days or more, with about 50% taking months or longer to discover. Once discovered though, we see fairly good reaction time, with about half of the organizations taking days or less to respond and contain the incident. This is far better than the norm, which is typically weeks or longer. Figure 27. Top 10 discovery methods for financially motivated incidents within Web App Attacks (n=122) Ext - customer Ext - fraud detection Int - IT audit Ext - unrelated party Ext-lawenforcement Int - fraud detection Int - reported by user Ext - actor disclosure Ext - audit Ext - monitor service 2% 2% 2% 2% 1% 1% 74% 3% 6% Total External Total Internal 88% 12% 4% COMPARING ATTACKS TO PATCH TIMEFRAMES Wouldn’t it be great to know that (quickly) patching web application vulnerabilities helps? This year we partnered with WhiteHat Security in order to combine and compare the incident data we’ve collected against the vulnerability assessment data they collect from tens of thousands of websites across hundreds of the most well-known organizations. After some back and forth, we decided first to break out the data by industries (because patterns emerge across industries), then we decided to compare two data points on web vulnerabilities to the incident data: the average (mean) vulnerabilities per site and median days to patch. We assumed that industries with fewer vulnerabilities and quicker patch time would be less represented in the breach data (i.e., have fewer incidents) and so we applied some good old-fashioned statistics and were admittedly let down when we didn’t see the relationship4 we were expecting. What we found is a non-finding and the only valid conclusion to draw from this is that more work is needed to understand the relationship between web application vulnerabilities and security incidents. With a non-finding, we can only speculate on why we are seeing these results. Perhaps this is telling us that no industry is doing enough. We know three out of four web-based compromises occur in hours or less of first contact, and maybe fixing vulnerabilities in 10 days versus 70 days doesn’t help all that much. Plus, the attacker only exploits one (maybe two) vulnerabilities. But a different explanation could be that our lens was focused too wide, and we could learn more by matching the high-quality WhiteHat data with specific incident data within the same sample. Whatever the causes, we do know that web application attacks occur often enough to repeat what is said in the WhiteHat Website Security Statistics Report,5 “What’s needed is more secure software, NOT more security software.” WEBAPP ATTACKS 21VERIZON 2014 DATA BREACH INVESTIGATIONS REPORT
  24. 24. POINT-OF-SALE INTRUSIONS WEBAPP ATTACKS INSIDERAND PRIVILEGEMISUSE PHYSICALTHEFT ANDLOSS MISCELLANEOUS ERRORS CRIMEWARE PAYMENTCARD SKIMMERS CYBER- ESPIONAGE DOS ATTACKS EVERYTHING ELSE RECOMMENDED CONTROLS Single-password fail The writing’s on the wall for single-factor, password-based authentication on anything Internet-facing. Even though it may draw you out of a known comfort zone, if you’re defending a web application seek out alternatives to this method of identity verification. If you’re a vendor in the web application space, also consider mandating alternative authentication mechanism for your customers. Rethink CMS And we mean “rethink” in every way. If you’re committed to an active platform (Joomla!, Drupal, WordPress, etc.), then set up an automated patch process. If an automated patch process isn’t viable, then develop a manual process and stick to it. This is especially true for the third-party plugins. Another way to rethink CMS is to consider a static CMS framework. Instead of executing code for every request and generating the content, static CMS will pre-generate those same pages, removing the need to execute code on the server for every request. Validate inputs Even though we’ve been tilting at this windmill for years, the advice still holds true. The best way to be sure your web application won’t be exploited is to seek out and fix the vulnerabilities before the attackers do (and they will). If you don’t have access to the source code and/or the developers, be sure to have something in place (e.g., a contract) to fix the problems when they’re found. Enforce lockout policies Brute force attacks aren’t the leading method in this section, but they’re still worthy of mention. By instituting counter- measures, such as a slowing down the rate of repeated attempts or temporarily locking accounts with multiple failed attempts, the rate of successful brute force attempts will more than likely dissipate and disappear (although you may still be dealing with that pesky bot poking at your accounts every now and then). Monitor outbound connections While many web-based attacks rely heavily on the existing firewall bypass protocol (HTTP), many others change the victim’s web server into a client. Critical points in the attack chain are pulling additional malware to continue the attack, exfiltrating compromised data, or attacking others on command. So unless your server has a legitimate reason to send your data to Eastern Europe or DoS’ing others is part of the business plan, try to lock down your web server’s ability to do so. Figure 29. Timespan of events within Web App Attacks Discoveryn=70Containmentn=41 0% 0% 3% 2% 11% 17% 16% 11% 0% 5% 42% 22% 29% 41% Compromisen=43Exfilterationn=33 0% 0% 0% 0% 3% 27% 21% 21% 18% 9% 19% 42% 12% 23% 0% 0% 0% 5% Hours Days Weeks Months Years Never Seconds Minutes Figure 28. Top 5 discovery methods for ideologically motivated incidents within Web App Attacks (n=775) Ext - unrelated party Ext - fraud detection Ext-lawenforcement Int-reportedbyuser Ext - actor disclosure Ext - customer Int - unknown Int-antivirus Int - fraud detection Other <1% <1% <1% <1% <1% <1% 93% <1% 4% Total External Total Internal 98% 2% 1% WEBAPP ATTACKS 22 VERIZON ENTERPRISE SOLUTIONS
  25. 25. POINT-OF-SALE INTRUSIONS WEBAPP ATTACKS INSIDERAND PRIVILEGEMISUSE PHYSICALTHEFT ANDLOSS MISCELLANEOUS ERRORS CRIMEWARE PAYMENTCARD SKIMMERS CYBER- ESPIONAGE DOS ATTACKS EVERYTHING ELSE INSIDER AND PRIVILEGE MISUSE An organization’s intellectual property is among its most valuable assets, frequently driving its ability to compete in the market. In many cases, organizations also have custody of vast amounts of data about the customers they serve, the employees who serve them, and the relationships they rely upon to do business. This data has value to the organization, but also to those who would seek it for their own personal benefit or myriad other reasons. For the misuse pattern, we focus on those who already have a trusted place inside the organization. Arguably, the most prominent case of internal misuse in the headlines this past year has been that of U.S. government contractor Edward Snowden. While this is an extreme example of the damage that determined insiders can inflict, it illustrates the risk that exists when an organization must place trust in individuals. Figure 30 lists the top threat actions observed across incidents fitting the misuse pattern. Note that not all are within the misuse category [mis]; stay tuned for more on that later. Not unexpectedly, privilege abuse — taking advantage of the system access privileges granted by an employer and using them to commit nefarious acts — tops the list. We realize that encompasses a very broad range of activities, but the overall theme and lesson differ little: most insider misuse occurs within the boundaries of trust necessary to perform normal duties. That’s what makes it so difficult to prevent. Remember that action varieties in VERIS are not mutually exclusive, and it’s common to see more than one in a single incident. Unapproved hardware and email misuse/data mishandling (tied) round out the top three actions in the misuse category, but they’re more a function of how the data is exfiltrated rather than how it’s acquired. Unapproved hardware refers to employees using devices like USB drives that are either forbidden altogether or allowed but subject to various restrictions. An employee sending intellectual property out to his or her personal address is an example of email misuse. We also reviewed cases where system administrators abused the email system, posing as another user and sending messages under that identity, with the goal of getting them fired. Data mishandling takes place when someone uses data in a manner counter to the organization’s policies. For example, a call center employee who writes customer credit card numbers down on paper, or an engineer who skirts policy by taking restricted documents home to examine on a personal computer. AT A GLANCE Description All incidents tagged with the action category of Misuse — any unapproved or malicious use of organizational resources — fall within this pattern. This is mainly insider misuse, but outsiders (due to collusion) and partners (because they are granted privileges) show up as well. Top industries Public, Real Estate, Administrative, Transportation, Manufacturing, Mining Frequency 11,698 total incidents 112 with confirmed data disclosure Key findings Most crimes by trusted parties are perpetrated for financial or personal gain. The most noticeable shifts in the 2013 dataset, however, were an increase in insider espionage targeting internal data and trade secrets, and a broader range of tactics. We say “2013 dataset” because we do not believe the actual rate of such crimes increased significantly; we’re seeing the benefit of increased visibility from insider-focused partners. Privilege abuse [mis] Unapproved hardware [mis] Bribery [soc] Email misuse [mis] Datamishanding[mis] Useofstolencreds[hac] Unapproved workaround [mis] Theft [phy] Unapproved software [mis] Embezzlement [mis] 11% 11% 7% 5% 4% 4% 4% 88% 18% 16% Figure 30. Top 10 threat action varieties within Insider Misuse (n=153) INSIDERAND PRIVILEGEMISUSE 23VERIZON 2014 DATA BREACH INVESTIGATIONS REPORT
  26. 26. POINT-OF-SALE INTRUSIONS WEBAPP ATTACKS INSIDERAND PRIVILEGEMISUSE PHYSICALTHEFT ANDLOSS MISCELLANEOUS ERRORS CRIMEWARE PAYMENTCARD SKIMMERS CYBER- ESPIONAGE DOS ATTACKS EVERYTHING ELSE With more incidents than ever before involving trusted parties, we could more easily see how they go about acquiring the data when their own access is insufficient. In addition to abusing entrusted privileges and resources, we observed hacking techniques to elevate privileges (often by stealing others’ credentials) and circumvent controls, various forms of social engineering, and the use of malware like keyloggers and backdoors. These cads have even resorted to physical theft, taking documents such as blueprints and other intellectual property, often denying availability to the original organization by taking the only copy. It’s also worth noting that the corporate LAN was the vector in 71% of these incidents, and 28% took advantage of physical access within the corporate facility. This means the majority of employees perpetrated their acts while in the office right under the noses of coworkers, rather than hopping through proxies from the relative safety of their house. If someone wants to use these statistics to loosen up work-at-home policies and tear down cube farms in favor of more open floor plans — you have our blessing. Let’s take a look at the people committing these crimes. While payment chain personnel and end-users were still prominent, managers (including those in the C-suite) came in higher than in prior years. You know the type; one of those straight shooters with upper management written all over him. They often have access to trade secrets and other data of interest to the competition and, tragically, are also more likely to be exempted from following security policies because of their privileged status in the company.6 One of those “white-collar resort prisons” won’t do for their ilk. As mentioned in the beginning of this section, insiders aren’t the only ones who misuse entrusted privileges and resources. Figure 33 gives an account of external and partner actors who directly or indirectly participated in incidents of misuse. Organized criminals bribe insiders to steal data for fraud schemes. Former employees exploit still active accounts or other holes known only to them. Competitors solicit intellectual property to gain business advantages. To mount a proper defense, organizations must take into account that such players are on the field. Nearly all misuse incidents prior to 2013 centered on obtaining information to use for fraud. As Figure 34 shows, we saw more insider espionage targeting internal organizational data and trade secrets than ever before. According to The Recover Report,7 published by one of our DBIR contributors, Mishcon de Reya, the two most common scenarios involve perpetrators taking the data to: • Start their own competing company (30%). • Help secure employment with a rival (65%). This kind of thing is certainly not new — it’s largely due to the addition of more contributors who have a view into this type of activity than ever before. So, whether it’s fraud-ready data sold on the quick to criminals or internal secrets eventually sold to a competitor, insider crime is still “all about the Benjamins, baby.” Figure 31. Vector for threat actions within Insider Misuse (n=123) LAN access Physical access Other Remote access Non-corporate 71% 28% 2% 1% 21% Figure 32. Top 10 varieties of internal actors within Insider Misuse (n=99) Cashier End-user Finance Manager Callcenter Executive Other Developer System admin Auditor 9% 7% 7% 6% 6% 1% 23% 13% 17% 13% Figure 33. Variety of external actors within Insider Misuse (n=25) Organized crime Former employee Competitor Unaffiliated Acquaintance 36% 24% 8% 24% 16% Figure 34. Actor motives within Insider Misuse (n=125) 72%Financial Espionage Convenience Grudge Fun N/A 10% 3% 2% 4% 18% INSIDERAND PRIVILEGEMISUSE 24 VERIZON ENTERPRISE SOLUTIONS
  27. 27. POINT-OF-SALE INTRUSIONS WEBAPP ATTACKS INSIDERAND PRIVILEGEMISUSE PHYSICALTHEFT ANDLOSS MISCELLANEOUS ERRORS CRIMEWARE PAYMENTCARD SKIMMERS CYBER- ESPIONAGE DOS ATTACKS EVERYTHING ELSE Desktops are the most frequently compromised asset in this pattern, which makes sense because desktop computers are an employee’s primary interface to the rest of the network (Figure 36). Typically, this is where the data is stored, uploaded, or emailed out of the organization, or copied onto removable media. Databases and file servers, both repositories of so much valuable information, are also targeted regularly. Payment cards doesn’t refer to the variety of data, but rather actual cards that were run through handheld skimming devices (or otherwise copied) in the classic “evil waiter” scenario. As far as asset ownership, we see insiders abusing corporate-owned rather than employee-owned (“BYOD”) assets allowed for corporate use. However, we do see evidence they often leverage unapproved personal devices to help them get the data out of the organization (which shows up as use of unapproved hardware). Discovery methods for the majority of breaches have traditionally been dominated by external signals. For insider misuse, however, internal methods (55%) are responsible for detecting more incidents than external methods (45%). The most common way organizations detected insider crimes was when employees reported them. Discoveries triggered by financial and IT audits were also very common. Reviewing the books on Monday morning is an example of the former, and a promising example of the latter is a regular process to review access for exiting employees. The CERT Insider Threat Center (another partner of ours) focuses research on insider breaches, and it determined that in more than 70% of the IP theft cases, insiders stole the information within 30 days of announcing their resignation.8 On quite a few occasions, a review of the activity of outgoing employees with access to sensitive information allowed impacted organizations to detect the incident and act quickly to retrieve the information (hopefully before irreparable damage had been done). Figure 35. Variety of at-risk data within Insider Misuse (n=108) Personal Payment Internal Secrets Bank Credentials Medical Unknown Other Classified 3% 3% 2% 1% 34% 29% 27% 18% 14% 9% Figure 36. Top 10 assets affected within Insider Misuse (n=142) Desktop (user dev) Database (server) Other (server) Payment card (media) Cashier (people) File (people) Other (people) Web application (server) Unknown Laptop (user dev) 10% 9% 8% 8% 6% 5% 26% 12% 25% 22% Figure 37. Top 10 discovery methods within Insider Misuse (n=122) Ext - customer Int - reported by user Int - unknown Int - financial audit Int - IT audit Ext - fraud detection Ext - unknown Ext - law enforcement Int - fraud detection Ext - audit 9% 8% 6% 6% 6% 3% 16% 10% 13% Total External Total Internal 45% 55% 11% INSIDERAND PRIVILEGEMISUSE 25VERIZON 2014 DATA BREACH INVESTIGATIONS REPORT
  28. 28. POINT-OF-SALE INTRUSIONS WEBAPP ATTACKS INSIDERAND PRIVILEGEMISUSE PHYSICALTHEFT ANDLOSS MISCELLANEOUS ERRORS CRIMEWARE PAYMENTCARD SKIMMERS CYBER- ESPIONAGE DOS ATTACKS EVERYTHING ELSE Note the discovery timeline for misuse in Figure 38 — it looks very different from the overall timeline we see for other types of incidents. The majority of the misuse incidents were detected within days (which is great), but there’s also a not insignificant number of incidents (70 to be exact) that took years to discover (which ain’t so great). RECOMMENDED CONTROLS The root cause of data theft and other illicit acts by trusted parties is, rather obviously, an employee breaking bad. No, not in a Walter White sense; more like a white collar crime sense (though Walter did ***SPOILER ALERT*** murder his boss and execute a forced takeover of the business, so it is rather apropos). While it’s impossible to stop all rogue employees, there are some steps that can reduce the likelihood of an incident occurring, or at least increase your chances of catching it quickly. Know your data and who has access to it The first step in protecting your data is in knowing where it is, and who has access to it. From this, build controls to protect it and detect misuse. It won’t prevent determined insiders (because they have access to it already), but there are many other benefits that warrant doing it. Review user accounts Having identified the positions with access to sensitive data, implement a process to review account activity when those employees give notice or have been released. Disable user accounts as soon as an employee leaves the company (and, if warranted, before that). This has proven successful in either preventing the data from leaving the organization, or in retrieving it quickly to contain the incident. Watch for data exfiltration In the top misuse varieties, we see actions that facilitate the data transfer out of the organization — these are excellent places to set up controls to detect this type of activity. Many data loss prevention products cover the most common actions taken to steal sensitive information, and these are certainly worth exploring. Publish audit results From an awareness perspective, regularly publish anonymized results of audits of access. Let employees know that there are consequences and that the policies are being enforced. This can act as a powerful deterrent to bad behavior. Hours Days Weeks Months Years Seconds Minutes 32% 7% 28% 22% 8% <1% 4% Figure 38. Discovery timeline within Insider Misuse (n=1,017) INSIDERAND PRIVILEGEMISUSE 26 VERIZON ENTERPRISE SOLUTIONS
  29. 29. POINT-OF-SALE INTRUSIONS WEBAPP ATTACKS INSIDERAND PRIVILEGEMISUSE PHYSICALTHEFT ANDLOSS MISCELLANEOUS ERRORS CRIMEWARE PAYMENTCARD SKIMMERS CYBER- ESPIONAGE DOS ATTACKS EVERYTHING ELSE PHYSICALTHEFTAND LOSS To be honest, we debated whether or not to include a section on lost and stolen assets in this report. We decided, however, that we simply couldn’t ignore the blatant fact that such incidents — while not sexy or “cyber-y” — are among the most common causes of data loss/exposure reported by organizations. This is especially apparent in industries like Healthcare, where the disclosure of all incidents that potentially expose sensitive data is mandatory. And if there’s anything we know to be true about human nature, it’s that losing things and stealing things seem to be inherent predispositions. Studying the findings yielded a few interesting observations that may help inform practice, and that’s where we’ll focus our attention in this section. As we begin, keep in mind that we’re specifically talking about information assets;10 whatever was lost or stolen had to store, process, or transmit information in order to get our attention. Observation #1 relates to demographics; we have evidence that every type and size of organization loses stuff and/or has stuff stolen. That may not be much of a shock, but it’s at least noteworthy that this is the only incident pattern that applies across the board. Even farmers have problems with people that come and try to snatch their crops laptops. Speaking of laptops, they’re the most common variety of asset reported with this pattern. Incident reports — especially to CSIRTs — often don’t specify the asset lost or stolen. Thus, “some kind of user device” is all we can infer and explains why “Other (user dev)” is so frequent. Beyond that, it’s what you’d expect: computers, documents, and drives. The next thing to note is the ratio of loss to theft; losing information assets happens way more than theft, by a 15-to-one difference. And that’s important because it suggests the vast majority of incidents in this pattern are not due to malicious or intentional actions. Thus, the primary challenge is to a) keep employees from losing things (not gonna happen) or b) minimize the impact when they do. The smart money is on option b, though bio-implanted computing devices do hold some future promise for option a. That’s about all we’re going to say about loss, but theft still has a few more lessons for us. AT A GLANCE Description Pretty much what it sounds like — any incident where an information asset went missing, whether through misplacement or malice. Top industries Healthcare, Public, Mining Frequency 9,704 total incidents9 116 with confirmed data disclosure Key findings Loss is reported more often than theft. In a surprising finding, we discovered that assets are stolen from corporate offices more often than personal vehicles or residences. And while personal and medical information is commonly exposed, most losses/thefts are reported because of mandatory disclosure regulations rather than because of fraud. Other (user dev) Laptop (user dev) Documents (media) Desktop (user dev) Flash drive (media) Disk drive (media) Tapes (media) Other (server) Other (media) Database (server) 108 102 37 36 27 12 11 892 308 140 Figure 39. Top 10 action varieties of Theft/Loss (n=9,678) Figure 40. Top 10 locations for theft within Theft/Loss (n=332) Victim work area Personal vehicle Personal residence Victim secure area Partner facility Partner vehicle Public facility Victim grounds Public vehicle Victim public area 5% 4% 4% 3% 2% 2% 2% 43% 23% 10% PHYSICALTHEFT ANDLOSS 27VERIZON 2014 DATA BREACH INVESTIGATIONS REPORT
  30. 30. POINT-OF-SALE INTRUSIONS WEBAPP ATTACKS INSIDERAND PRIVILEGEMISUSE PHYSICALTHEFT ANDLOSS MISCELLANEOUS ERRORS CRIMEWARE PAYMENTCARD SKIMMERS CYBER- ESPIONAGE DOS ATTACKS EVERYTHING ELSE We find it quite surprising that the highest proportion of thefts occur in the victim’s work area, which basically refers to the main office space or cube farm (Figure 40). That suggests simply having sensitive information “behind locked doors” isn’t enough; there are still a lot of people inside those locked doors.11 Notice that thefts in internal high security areas are much less common, but still post higher than public facilities. That last bit is counterintuitive to the point of irrationality; we can’t help but suspect that people whose laptops are stolen when they take a potty break at the coffee shop simply report them as “lost” to save face. Personal residences and personal/partner/public vehicles serve as the venue for nearly 40% of thefts and remind us that devices on the go are prone to go missing. While it’s usually not known/reported exactly how actors gained physical access to these locations, over 80% of thefts where we do have that info involved disabling or bypassing controls. The remainder had access already, either because they were granted privileges or because it was a publicly accessible location. The final set of observations covers the variety of data that was compromised or, more often, potentially exposed when assets were lost or stolen. It’s worth pointing out that the primary reason most of these incidents are included is because they tripped some kind of mandatory reporting/disclosure requirement. The asset went missing, was determined to contain regulated information that is now exposed to potential unauthorized access, and therefore had to be reported. This explains the predominance of regulated data like personal or identifying information and medical records in Figure 42. RECOMMENDED CONTROLS The primary root cause of incidents in this pattern is carelessness of one degree or another. Accidents happen. People lose stuff. People steal stuff. And that’s never going to change. But there are a few things you can do to mitigate that risk. Encrypt devices Considering the high frequency of lost assets, encryption is as close to a no-brainer solution as it gets for this incident pattern. Sure, the asset is still missing, but at least it will save a lot of worry, embarrassment, and potential lawsuits by simply being able to say the information within it was protected. Also, periodically checking to ensure encryption is still active is right up there too. This will come in handy when the auditor or regulator asks that dreaded question: “How do you know for sure it was encrypted?” Keep it with you Encourage employees to keep sensitive devices in their possession and in sight at all times. Yes, this applies to fancy client dinners and visits to the restroom. It’s not a bad principle to apply to mobile devices in a corporate setting either. It may be awkward, but it’s safer than leaving it in the car or unattended in a room full of strangers. If it absolutely must be left in the car, lock it in the trunk before you leave the office and don’t leave it there overnight. Back it up Regular (and preferably automatic) backups serve a threefold purpose. They salvage weeks/months/years’ worth of irrecoverable work, get you productive again on a new device with minimal down time, and help establish what data was on the device to determine if disclosure is necessary. Lock it down In light of the evidence that so many thefts occur in the office, cabling or otherwise securing equipment to immovable fixtures should at least be considered. The big caveat, however, is that the majority of such thefts were documents taken from the filing cabinet and mobile devices (including laptops). A more effective strategy would be to move highly sensitive or valuable assets to a separate, secure area and make sure they stay there. BONUS - Use unappealing tech Yes, it’s unorthodox as far as recommendations go, but it might actually be an effective theft deterrent (though it will probably increase loss frequency). That shiny new MacBook Air on the passenger seat may be too tempting for anyone to resist, but only those truly dedicated crooks will risk incarceration for a 4” thick mid-90s lap brick. Or, if being the fastest hunk of junk in the galaxy is a must, perhaps there’s a lucrative aftermarket for clunky laptop covers. She may not look like much, but she’s got it where it counts, kid. Figure 42. Variety of at-risk data within Theft/Loss (n=3,824) Personal Medical Payment Internal Bank Classified Credentials Unknown Secrets Other 21 20 5 5 4 3 1 3,393 508 23 Figure 41. Vector of physical access within Theft/Loss (n=158) Disabled controls Bypassed controls Privileged access Uncontrolledlocation 60% 26% 11% 3% PHYSICALTHEFT ANDLOSS 28 VERIZON ENTERPRISE SOLUTIONS
  31. 31. POINT-OF-SALE INTRUSIONS WEBAPP ATTACKS INSIDERAND PRIVILEGEMISUSE PHYSICALTHEFT ANDLOSS MISCELLANEOUS ERRORS CRIMEWARE PAYMENTCARD SKIMMERS CYBER- ESPIONAGE DOS ATTACKS EVERYTHING ELSE MISCELLANEOUS ERRORS Nearly every incident involves some element of human error. For example, failing to apply a WordPress patch certainly leaves the application vulnerable to attack, but it doesn’t directly compromise the system. Some other threat actor/action is required to do that. Without drawing that distinction, this category would be so bloated with “incidents” that it would be difficult to extract useful information. It’s also worth noting that this pattern doesn’t include every incident in the Error category. Loss is a type of error, but we grouped it with theft (under Physical) in a different pattern because they share certain similarities (you no longer possess the device) and because it’s often difficult to determine loss vs. theft. Please keep this in mind as you view the top actions and assets in this section. Misdelivery (sending paper documents or emails to the wrong recipient) is the most frequently seen error resulting in data disclosure. There are only two difficult problems in computer science: cache invalidation, naming things, and off-by-one errors. Misdelivery (sending paper documents or emails to the wrong recipient) is the most frequently seen error resulting in data disclosure. One of the more common examples is a mass mailing where the documents and envelopes are out of sync (off-by-one) and sensitive documents are sent to the wrong recipient. A mundane blunder, yes, but one that very often exposes data to unauthorized parties. Misdelivery Publishing error Disposal error Misconfiguration Malfunction Programming error Gaffe Omission Other Maintenance error 6% 3% 3% 1% 1% 1% <1% 44% 22% 20% Figure 43. Top 10 threat action varieties within Miscellaneous Errors (n=558) AT A GLANCE Description Incidents where unintentional actions directly compromised a security attribute of an information asset. This does not include lost devices, which is grouped with theft instead. Top industries Public, Administrative, Healthcare Frequency 16,554 total incidents12 412 with confirmed data disclosure Key findings After scrutinizing 16K incidents, we’ve made a startling discovery — people screw up sometimes. (Nobel Prize, here we come!) The data seems to suggest that highly repetitive and mundane business processes involving sensitive info are particularly error prone. It’s also noteworthy that this pattern contains more incidents caused by business partners than any other. Documents (media) Web application (server) Desktop (user dev) File (server) Database (server) Other (server) Mail (server) Disk media (media) Disk drive (media) Other (media) 5% 5% 4% 3% 1% 1% 7% 9% 14% 49% Figure 44. Top 10 assets affected within Miscellaneous Errors (n=546) MISCELLANEOUS ERRORS 29VERIZON 2014 DATA BREACH INVESTIGATIONS REPORT